A 0.2 V Window Feature-Driven Cascaded Machine Learning Pipeline for Joint State of Health and Remaining Useful Life Estimation of Lithium-Ion Batteries

Le résumé fourni par la source

Machine learning methods have been demonstrated to develop health management strategies for lithium-ion batteries, such as state of health (SOH) and remaining useful life (RUL) estimation. However, current methods face challenges like additional testing, complex but non-generalizable health indicators (HI), and data leakage. To address these problems, this study proposes a cascaded Temporal Convolutional Network (TCN) - Grey Wolf Optimizer (GWO) - Gaussian Process Regression (GPR) machine learning framework for accurate SOH and RUL joint estimation with a novel HI. Initially, the electrochemical aging mechanism is investigated using incremental capacity analysis. Subsequently, a low-complexity, mechanism-included HI is extracted from the charging voltage with only a 0.2V range. Moreover, the TCN and Savitzky-Golay filter are utilized to predict future HI trajectories, avoiding the data leakage. The GPR with a novel combined covariance function is employed to estimate the SOH with the HI as inputs. Furthermore, the GWO is used to tune the hyperparameters of GPR, thereby forming an auto-adjusting SOH estimation model that predicts future SOH trajectories to obtain predicted RUL. The performance of the framework is validated on different batteries with mean absolute error and average relative error of only 4 cycles and 2.36%, respectively, demonstrating its accuracy and generalizability.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.